Research On EEG Data Analysis Method Based On Tunable Q-Factor Wavelet Transform And Transfer Learning

At present,clinically,the detection of epileptic seizure and the staging of different stages of sleep are mainly completed by professional doctors taking visual inspections of patients’ EEG.Among them,epilepsy is a very serious nervous system caused by the rapid discharge of neurological disease.When the onset of disease occurs,the patient’s brain will have a short-term dysfunction,which will lead to symptoms such as cognitive decline,uncontrolled limbs,and even death of the patient,which has a huge impact on the normal life of the patient.Therefore,timely and effective detection of epileptic seizure becomes particularly important for the diagnosis and treatment of epilepsy patients.Sleep,as a basic physiological activity of the human body,plays an irreplaceable role in the recovery and consolidation of the body.However,with the increase of social pressure,people’s sleep quality continues to decline,and sleep-related diseases are more and more torturing people’s body and mind.This situation has increased people’s demands for improving sleep quality.When entering a sleep state,the brain loses consciousness,and its response to external stimuli is reduced.Diseases that are not easily exposed when awake will appear in different sleep stages.Therefore,the sleep staging can not only achieve an in-depth understanding of the various stages of sleep,but also the diagnosis and research of sleep disorders-related diseases.Although the visual detection of epilepsy and sleep EEG signals by professional doctors can also achieve the purpose of detection,these process are often time-consuming and labor-intensive due to the large amount of EEG signal data,and they are also prone to misjudgment.Therefore,there is an urgent need to design a scheme for automatic detection of EEG signals based on seizure and sleep staging.Based on the single-channel EEG signal,this paper proposes a new analysis method to realize the automatic detection of epileptic seizure and sleep staging.First,a new data representation method is proposed for EEG signals.The subbands after the Tunable Q factor wavelet transform are partially reconstructed and rearranged according to the average energy ratio threshold,and the rearranged three-channel data is expressed as two multi-dimensional color image data;further combining the existing EEG automatic detection algorithm and the research of the depth separable convolutional network Xception model,using the fully trained pre-training model parameters on the Image Net data set to initialize the network parameters;finally,through the transfer learning method,the Xception model is fine-tuned to adapt to the tasks of automatic seizure detection and sleep staging.Meanwhile,the effectiveness of the raised approach above can be sufficiently verified through numerous experiments.Compared with the existing methods,the automatic detection methods proposed in this paper have obtained better detection results,without manual design and feature extraction,and have good practicability.